Career Outlook

Is software engineering still worth it? What the numbers actually show

This question gets asked in two very different tones. Someone deciding whether to spend four years and a lot of money on a computer science degree is asking one thing. Someone eight years into the job watching their team shrink is asking something else. The honest answer is different for each of them, and most of what gets written about this topic serves neither.

So here is the data, with sources, and then what it means depending on where you're standing.

The short version

Software engineering as a profession is not dying. By every employment measure that exists, it's large, well paid, and projected to keep growing.

What broke is the entrance. Getting your first engineering job is dramatically harder than it was five years ago, and that shift is real, measurable, and worse than the headline numbers suggest. The risk isn't that the career disappears. It's that the bottom rung of the ladder has been sawed off, and almost everything written about this conflates the two.

If you're already in the field, you're probably worried about the wrong thing. If you're trying to get in, it's harder than anyone is telling you, and there are specific things that still work.

What the employment data says

The Bureau of Labor Statistics counts about 1.9 million people employed as software developers, quality assurance analysts, and testers as of 2024. It projects 15 percent growth through 2034, which it classifies as much faster than average, and roughly 129,200 openings a year over that decade. Median pay for software developers specifically is about $133,000.

Unemployment tells a similar story. As of June 2026, the unemployment rate for tech occupations sat around 2.9 percent against a national rate of 4.2 percent. People working in the field are, on the whole, still employed.

It's worth knowing that BLS splits this into separate categories, because that split is the source of a lot of confusion. "Computer programmer" is a narrower, older classification covering implementation work, and it's shrinking fast: about 121,200 people, projected to decline 6 percent. "Software developer" is the broader role that includes design and requirements, and it's the one that's growing. Headlines about programmer employment hitting its lowest level in decades are describing the small shrinking category, not the large growing one. That distinction gets flattened constantly and it makes the picture look far worse than it is.

What the hiring data says, which is different

Employment and hiring are not the same thing, and this is where the picture gets genuinely worse. You can have high employment and a frozen market at the same time, which is roughly what's happening.

Indeed's Hiring Lab tracks job postings against a February 2020 baseline. Software development postings peaked in early 2022 at more than double that baseline, then fell hard. As of April 2026 the index sat around 30 percent below where it was before the pandemic. So postings are well under pre-2020 levels, and dramatically under the 2022 peak that a lot of people still use as their mental reference point.

The recent direction is up, though, and that matters. Postings were up 14 percent year over year as of April 2026, and roughly 15 percent higher than in February 2025. The bottom appears to have been in 2024 and 2025. This is not a market in free fall. It's a market that fell a long way and has started climbing back from a low base.

The competition inside that market got much steeper regardless. Recruiting data from Ashby, covering more than 100 million applications, shows applications per hire have roughly tripled since 2021 and now average over 300. A candidate is about half as likely to get an interview as five years ago. That's across role types, not software alone, but it describes the experience accurately: the same application volume produces a fraction of the responses it used to.

Where it actually broke

Everything above is the aggregate. Split it by experience level and the aggregate stops being useful, because the damage is concentrated almost entirely at the entry point.

SignalFire, which tracks hiring across hundreds of millions of professional profiles, found new graduate hiring at the largest tech companies down roughly 65 percent compared to 2019, and down about 76 percent at early-stage startups. Graduates of top-20 computer science programs in 2025 were about 45 percent less likely to take an engineering job at a major tech company than they were a few years earlier. New grads are now roughly twice as likely to start a company as they were at the 2022 peak, which is less a sign of ambition than of what happens when the normal path closes.

New York Fed data on recent college graduates puts unemployment for computer science majors at 7.0 percent and computer engineering at 7.8 percent, against 5.6 percent for recent graduates overall. Computer science graduates have higher unemployment than the average major. That is a genuinely new development and it deserves to be taken seriously rather than explained away.

The most careful evidence comes from Stanford's Digital Economy Lab, which used payroll records from the largest US payroll processor rather than surveys. They found workers aged 22 to 25 in the most AI-exposed occupations showed a 16 percent relative decline in employment, while more experienced workers in those same occupations stayed flat or grew. The adjustment showed up in hiring, not in wages. Companies didn't cut pay. They stopped hiring juniors.

That's the shape of it. The field is fine. The on-ramp is not.

About AI writing all the code

This is where the discussion usually goes off the rails, so it's worth separating what's been measured from what's been claimed.

On the claims side: Microsoft's CEO said in April 2025 that 20 to 30 percent of code in the company's repositories was AI-written. Google's CEO said more than 25 percent of new code was AI-generated in late 2024, then well over 30 percent by spring 2025, and the reported figure has kept climbing since. None of these disclose a methodology. Nobody says whether an accepted autocomplete suggestion counts the same as a function written end to end, and the numbers are self-reported with no outside audit. TechCrunch, reporting the Microsoft figure, noted plainly that it's unclear how any of this is being measured.

Some of the loudest claims have been walked back by the people who made them. Salesforce announced in 2025 that it would hire no additional software engineers, citing AI productivity gains. By April 2026 its CEO was saying AI "cannot operate autonomously," that his engineering organization was maybe 30 percent more productive rather than 100 percent, and pointing at the fact that AI labs themselves are still hiring engineers as evidence the technology hasn't arrived where he'd said it had.

On the measured side, the results are stranger than either camp expected. METR ran a controlled trial with 16 experienced open source developers working on their own large, mature codebases across 246 real issues, randomly assigning whether they could use AI tools. The developers took 19 percent longer to finish when they used AI. They had predicted beforehand it would speed them up by 24 percent, and afterward they still believed it had sped them up by 20 percent. They were wrong about their own performance in the wrong direction. The authors are careful to say this doesn't generalize to all software work, and it shouldn't be treated as proof AI doesn't help. What it does show is that self-reported productivity gains on this topic are close to worthless as evidence.

Indeed's own labor economist has repeatedly made a point that cuts against the simple story: tech job postings fell in a pattern that closely tracks the broader economy, including sectors with almost no AI exposure, and nearly half the decline happened before ChatGPT was even released. Rates rose, pandemic overhiring corrected, and the timing of AI's arrival makes it a very convenient explanation for decisions that were already underway.

None of that means AI is irrelevant. It means the causal story is much less settled than a headline suggests, and anyone telling you confidently that AI did or didn't cause this is ahead of the evidence.

The part that should actually worry a junior engineer

There's one finding here that gets less attention than it deserves, and it's not about jobs at all.

Anthropic ran a trial with 52 junior engineers working with a Python library none of them knew, splitting them into a group that used AI assistance and a group that wrote it by hand. Afterward, both groups took a comprehension test covering debugging, reading code, writing code, and conceptual understanding. The AI-assisted group scored 50 percent. The hand-coding group scored 67 percent. The gap was largest on debugging. And the AI group finished only about two minutes faster, a difference small enough to be statistical noise.

The detail that matters most: engineers who used AI to ask questions and understand what they were doing kept their comprehension. Engineers who used it to generate and fix without engaging were the fastest and scored the worst.

Put that next to the hiring data and you get the real problem. The tasks that used to be how juniors learned, the tedious, unglamorous, slightly-too-hard work, are exactly the tasks AI does well. So the work that built competence is being automated at the same moment companies stopped hiring the people who needed it. That's a harder problem than a bad hiring year, because it compounds.

If you're deciding whether to start

Going in with your eyes open beats going in optimistic. The honest framing is that this is a good career with a bad entrance, and you need a plan for the entrance specifically rather than assuming a degree produces a job.

Assume the first job takes longer than you were told. Budget a year of searching, not three months, and make sure your finances and your expectations can survive that. Most of the despair in this field right now comes from people who planned for a market that stopped existing in 2022.

Be deliberate about how you use AI while you're learning. Based on the Anthropic result, the difference between using it well and using it badly is whether you engage with what it produces or just accept it. Ask it why. Make it explain the thing it wrote. Write the hard parts yourself on purpose, especially debugging, because that's where the measured gap was widest and it's the skill interviews test hardest.

Build things that required judgment, and be ready to talk about the decisions rather than the features. When a hiring manager can assume any candidate can generate working code, what separates people is whether they can explain why this approach and not that one. That's an interview skill and it's built by making real decisions, not by finishing tutorials.

Go where juniors still get hired. The freeze isn't uniform. Large tech companies and venture-backed startups cut new grad hiring hardest. Non-tech companies with real engineering needs, government and defense contractors, healthcare systems, insurers, logistics, and manufacturers, are far less represented in the layoff data and far less likely to have an AI-driven hiring philosophy. The work is less glamorous and the first job matters more than the logo on it.

If you're already in

Your risk profile is close to the opposite of what the discourse suggests. The data shows experienced engineers holding steady while entry-level hiring collapsed. You are not the group being displaced.

What has changed is that your position is much less liquid than it was. Postings are still well below pre-pandemic levels and applications per opening have tripled, so leaving voluntarily is a bigger decision than it was in 2021. That's an argument for keeping your options warm rather than for panic. Know what you'd be worth, keep the relationships alive, and don't let a search start from zero.

The genuine risk isn't that AI replaces you. It's that the work drifts toward reviewing and directing generated code, and the skills that made you valuable quietly stop getting exercised. The METR result is worth sitting with here, since experienced developers were measurably worse off while feeling measurably better. Whatever your process is, verify it against something real rather than how it feels.

So: is it still worth it? For someone already working in it, yes, and the data isn't close. For someone deciding whether to start, it's worth it if you go in knowing the first job is the hard part and you plan for that specifically instead of assuming the pipeline will carry you.

What isn't worth doing is making this decision off headlines. The numbers about programmers collapsing describe a different job category. The numbers about AI writing most of the code come from companies that haven't said what they're counting. And the number that actually matters, the one about entry-level hiring falling by two thirds, barely gets mentioned at all.

Quick reference

Employed as software developers: About 1.9 million (BLS, 2024)
Projected growth: 15% through 2034, about 129,200 openings a year
Median pay: About $133,000 for software developers
Tech unemployment: 2.9% vs 4.2% nationally (June 2026)
Job postings: About 30% below Feb 2020, but up 14% year over year
New grad hiring: Down about 65% at big tech, 76% at startups vs 2019
CS grad unemployment: 7.0%, against 5.6% for all recent grads
Applications per hire: Over 300, roughly tripled since 2021